摘要 |
PROBLEM TO BE SOLVED: To rapidly retrieve an HMM (hidden Markov model) of high likelihood without omission. SOLUTION: Before retrieval processing, degeneracy processing is performed to the state of an HMM in a database and an output signal so as to reduce an integration error, and the resultant is stored in the database. If a retrieval request is acquired, the degenerated HMM is read from the database, approximation accuracy of every HMM is calculated, a solution candidate for retrieval is calculated, the likelihood of an original HMM stored in the database is calculated, a threshold of pruning is updated, and the HMM is pruned by approximate likelihood to conduct retrieval. COPYRIGHT: (C)2007,JPO&INPIT
|